Recommendation systems typically use historical interactions between users and items topredict what other items can be of interest to a user. The recommendations are based onpatterns in how users interact similarly with items. This thesis investigates if it is possible toimprove the quality of the recommendations by including more information about the items inthe model that predicts the recommendations. More specifically, the use of deep learning toextract information from item images is investigated. To do this, two types of collaborativefiltering models, based on historic interactions, are implemented. These models are thencompared to different collaborative filtering models that either make use of user and itemattributes, or images of the items. Three pre-trained image classification models are used toextract useful item features from the item images. The models are trained and evaluated using adataset of historic transactions and item images from the online sports shop Stadium, given bythe thesis supervisor. The results show no noticeable improvement in performance for themodels using the images compared to the models without images. The model using the userand item attributes performs the best, indicating that the collaborative filtering models can beimproved by giving it more information than just the historic interactions. Possible ways tofurther investigate using the image feature vectors in collaborative filtering models, as well asusing them to create better item attributes, are discussed and suggested for future work.

No description available.

In this thesis, we explore the idea of using binary visualization and image analysis to detect anomalous activity on an Industrial Internet of Things (IIoT) based network. The data is gathered into a pcap file and then fed into our encoder, which uses a space-filling curve to convert the 1-dimensional stream of data into pixels with a specific red, blue, and green gradient value.  The pixels create an image which is then given to an image analysis system based on a Convolutional Neural Network, which classifies if the traffic supplied is malicious or not. The results show that using a Binary and Multiclass classifier approach to the image analysis both work well reaching an accuracy of 100% and 94% respectively. While the binary classifier is more accurate both succeed at separating Malicious from Benign traffic. The choice of space-filling curves in our binary visualization ended up having little to no impact on overall classification accuracy.

We study modifications to the novel stochastic watershed method for segmentation of digital images. This is a stochastic version of the original watershed method which is repeatedly realized in order to create a probability density function for the segmentation. The study is primarily done on synthetic images with both same-sized regions and differently sized regions, and at the end we apply our methods on two endothelial cell images of the human cornea. We find that, for same-sized regions, the seeds should be placed in a spaced grid instead of a random uniform distribution in order to yield a more accurate segmentation. When images with differently sized regions are being segmented, the seeds should be placed dependent on the gradient, and by also adding uniform or gaussian noise to the image in every iteration a satisfactory result is obtained.

Satellites commonly use onboard digital cameras, called star trackers. A star tracker determines the satellite's attitude, i.e. its orientation in space, by comparing star positions with databases of star patterns. In this thesis, I investigate the possibility of extending the functionality of star trackers to also detect the presence of resident space objects (RSO) orbiting the earth. RSO consist of both active satellites and orbital debris, such as inactive satellites, spent rocket stages and particles of different sizes. I implement and compare nine detection algorithms based on image analysis. The input is two hundred synthetic images, consisting of a portion of the night sky with added random Gaussian and banding noise. RSO, visible as faint lines in random positions, are added to half of the images. The algorithms are evaluated with respect to sensitivity (the true positive rate) and specificity (the true negative rate). Also, a difficulty metric encompassing execution times and computational complexity is used. The Laplacian of Gaussian algorithm outperforms the rest, with a sensitivity of 0.99, a specificity of 1 and a low difficulty. It is further tested to determine how its performance changes when varying parameters such as line length and noise strength. For high sensitivity, there is a lower limit in how faint the line can appear. Finally, I show that it is possible to use the extracted information to roughly estimate the orbit of the RSO. This can be accomplished using the Gaussian angles-only method. Three angular measurements of the RSO positions are needed, in addition to the times and the positions of the observer satellite. A computer architecture capable of image processing is needed for an onboard implementation of the method.

Cell fate is known to be triggered by cues from the extracellular matrix including its chemical, biological and physical characteristics. Specifically, mechanical and topological properties are increasingly recognized as important signals. The aim of this work was to provide an easy-accessible biomimetic in vitro platform of topologically defined collagen I matrices to dissect cell behaviour under various conditions in vitro. We reconstituted covalently bound layers of three-dimensional (3D) networks of collagen type I and collagen type V with a defined network topology. A new erosion algorithm enabled us to analyse the mean pore diameter and fibril content, while the mean fibril diameter was examined by an autocorrelation method. Different concentrations and ratios of collagen I and V resulted in pore diameters from 2.4 μm to 4.5 μm and fibril diameters from 0.6 to 0.8 μm. A comparison of telopeptide intact collagen I to telopeptide deficient collagen I revealed obvious differences in network structure. The good correlation of the topological data to measurements of network stiffness as well as invasion of human dermal fibroblasts proofed the topological analysis to provide meaningful measures of the functional characteristics of the reconstituted 3D collagen matrices.

Object-based image analysis (OBIA) has been increasingly used to identify terrain features of archaeological sites, but only recently to extract subsurface archaeological features from geophysical data. In this study, we use a semi-automated OBIA to identify Archaic (8000-1000 BC) hearths from Ground-Penetrating Radar (GPR) data collected at David Crockett Birthplace State Park in eastern Tennessee in the southeastern United States. The data were preprocessed using GPR-SLICE, Surfer, and Archaeofusion software, and amplitude depth slices were selected that contained anomalies ranging from 0.80 to 1.20 m below surface (BS). Next, the data were segmented within ESRI ArcMap GIS software using a global threshold and, after vectorization, classified using four attributes: area, perimeter, length-to-width ratio, and Circularity Index. The user-defined parameters were based on an excavated Archaic circular hearth found at a depth greater than one meter, which consisted of fire-cracked rock and had a diameter greater than one meter. These observations were in agreement with previous excavations of hearths at the site. Features that had a high probability of being Archaic hearths were further delineated by human interpretation from radargrams and then ground-truthed by auger testing. The semi-automated OBIA successfully predicted 15 probable Archaic hearths at depths ranging from 0.85 to 1.20 m BS. Observable spatial clustering of hearths may indicate episodes of seasonal occupation by small mobile groups during the Archaic Period.

Urban image analysis is one of the most important problems lying at the intersection of computer graphics and computer vision research. In addition, Convolutional Sparse Coding (CSC) is a well-established image representation model especially suited for image restoration tasks. This dissertation handles urban image analysis using an asset extraction framework, studies CSC for the reconstruction of both urban and general images using supervised data, and proposes a better computational approach to CSC. Our asset extraction framework uses object proposals which are currently used for increasing the computational efficiency of object detection. In this dissertation, we propose a novel adaptive pipeline for interleaving object proposals with object classification and use it as a formulation for asset detection. We first preprocess the images using a novel and efficient rectification technique. We then employ a particle filter approach to keep track of three priors, which guide proposed samples and get updated using classifier output. Tests performed on over 1000 urban images demonstrate that our rectification method is faster than existing methods without loss in quality, and that our interleaved proposal method outperforms current state-of-the-art. We further demonstrate that other methods can be improved by incorporating our interleaved proposals. We also extend the applicability of the CSC model by proposing a supervised approach to the problem, which aims at learning discriminative dictionaries instead of purely reconstructive ones. We incorporate a supervised regularization term into the traditional unsupervised CSC objective to encourage the final dictionary elements to be discriminative. Experimental results show that using supervised convolutional learning results in two key advantages. First, we learn more semantically relevant filters in the dictionary and second, we achieve improved image reconstruction on unseen data. We finally present two computational contributions to the state of the art in CSC. First, we significantly speed up the computation by proposing a new optimization framework that tackles the problem in the dual domain. Second, we extend the original formulation to higher dimensions in order to process a wider range of inputs, such as RGB images and videos. Our results show up to 20 times speedup compared to current state-of-the-art CSC solvers.

Retinoic acid, suspended in cottonseed oil, was administered via gavage to pregnant mice (ICR strain) on day 11 (E 11) of gestation at doses of either 20, 40, or 80 mg/kg. Fetuses were examined for external malformations on day 17 (E 17). Retinoic acid treatment induced micromelia (with the elimination of several long bones at higher doses) and digital defects (ectrodactyly and syndactyly) in a dose‐dependent manner in fetuses examined on day 17. Hindlimbs were affected more than forelimbs. In another group of experiments, limbs exposed to retinoic acid treatment in utero on E 11 were cultured on E 12 and maintained for 3 days in submerged culture. Cultured limbs were examined qualitatively for digital and long bone defects, and image analysis of the area and form of bone anlagen of cultured limbs was used to quantitatively evaluate the teratogenic potential of retinoic acid. The qualitative evaluation indicated that the retinoic acid‐induced effects obtained in vivo and with pretreated, cultured limbs were essentially the same, except that the severity of regional effects changed as a result of culture. The incidence of ectrodactyly was higher with cultured limbs than with E 17 fetal limbs, but fewer cultured limbs were missing long bones. These results suggest that culturing limbs, after they have been pretreated in utero, modifies their response to a teratogen and demonstrates that the paw skeleton is extremely sensitive to teratogen treatment under these experimental conditions. Therefore, care must be exercised when attempting to compare in vivo and in vitro teratogenic data. This study also clearly demonstrates the power and usefulness of image analysis for quantitative evaluation of both the area and form of a cultured specimen such as the developing limb bud. Quantitative, image analysis of cultured limbs showed a dose‐dependent decrease in area of both fore‐ and hindlimbs. The effect was most severe in hindlimbs. In the forelimb, the paw was affected more than the long bones; as the dose increased, this disparity of effect also increased. With the hindlimb, a greater effect on the paw occurred only at 80 mg/kg. Computing the soft tissue/bone ratio illustrated that retinoic acid had a greater effect on chondrogenic tissue than on soft tissue.

Forelimbs of Day 11 and Day 12 embryonic mice were excised and cultured for 3 d in the presence of either 0.25 μg (8×10-7 M), 0.5 μg(1.7×10-6 M), or 1.0 μg (3.3×10-6 M) of all-rans retinoic acid (RA) per milliliter of culture medium. Cultured limbs were fixed, stained, and mounted whole on glass slides and evaluated with computerized optical image analysis for RA-induced effects on the area and shape of the total limb and individual bone anlagen. Relative effects of RA on total bone, soft tissue, long bone, and paw regions were also examined. With Day 11 forelimbs total bone area was increased by 10.5% by the low dose of RA. The increase was mostly in long bones and at the expense of soft tissue. Total bone area was increased 9.3% with Day 12 forelimbs. This increase was primarily in the paw. The high dose of RA decreased Day 11 forelimb area, primarily affecting long bones. Day 12 forelimbs were not significantly affected by the high dose of RA. Effects of the imtermediate dose were primarily limited to reduction in soft tissue area. Long bone:paw and soft tissue: bone ratios reflected these effects. The high dose produced a consistent rounding or shortening of Day 11 forelimb bones. On Day 12 0.5 μg/ml RA produced an inconsistent pattern of rounding of bone anlagen. Treatment with the high dose on Day 12 produced angular rather than rounded contours in many cases, as indicated by shape factor values closer to zero than obtained with controls. These data show that direct exposure to RA can affect both the size and shape of bone anlagen of the developing limb; the low dose enhances and the high dose depresses development. The results support previous studies which suggest that RA may play a critical role in the control of cell activities such as cell migration, proliferation, and cytodifferentiation in the development of the cartilaginous bone anlagen.